This video features a series of presentations from a bioinformatics conference, showcasing new tools, workflows, and analysis strategies for complex biological data within the Galaxy and Bioconductor ecosystems. The talks cover: the Intergalactic Workflows Commission (IWC), a community for standardized Galaxy workflows; new Bioconductor packages for analyzing digital pathology images from TCGA using deep learning; the application of the Trident toolkit in Galaxy for histopathology image pre-processing and predicting drug response; advanced methods for analyzing high-resolution Visium HD spatial transcriptomics data at various scales; a structure-based spatial omics analysis approach focusing on multicellular anatomical structures using the Spatula package; and a Galaxy workflow designed to accelerate the manual curation of vertebrate genomes by integrating Hi-C maps with other genomic evidence.
A Showcase of Bioinformatics Tools for Complex Data Analysis
The IWC is a community-driven initiative to create a central, version-controlled repository of production-ready Galaxy workflows on GitHub.
Goal: To lower the barrier for workflow use, sharing, and development within the Galaxy community.
01:50 Finding Workflows: The main entry point is iwc.galaxy.org, which hosts over 100 workflows across various disciplines.
02:41 Workflow Features: Each workflow has its own page with a description, metadata, a DOI via Zenodo, a visual preview, and versioning through GitHub releases.
05:02 "Run with Sample Data": A key feature allows users to launch a workflow on a chosen Galaxy instance with pre-filled test data from Zenodo, simplifying execution.
06:06 Benefits of Contributing: Contributors gain discoverability, a DOI for their work, and access to automated testing and tool update suggestions.
This project focuses on analyzing H&E (hematoxylin and eosin) stained digital pathology images from The Cancer Genome Atlas (TCGA) dataset.
11:52 Data Scale: The team processed 11,764 diagnostic images, totaling 11 TB of data.
13:05 Feature Extraction: Deep learning models like Hover-Net are used for nucleus segmentation and classification (e.g., neoplastic, stroma), followed by feature extraction using squidpy and scikit-image.
14:52imageTCGA Package: The extracted features and analysis tools are being integrated into a new Bioconductor package called imageTCGA, which includes a Shiny app for data exploration.
17:12 Huge Time Savings: The entire analysis pipeline is estimated to take over two years of computation, but the final package will make the results accessible in minutes.
This talk discusses developing tools for applying deep learning models to histopathology images within Galaxy.
21:53 Importance of Pre-processing: The speaker highlights the need to handle image artifacts, magnification differences, and stain variability before model training.
22:33 The Trident Toolkit: The team uses the Trident toolkit for pre-processing, including tissue segmentation and patch extraction. It is being implemented as a Galaxy tool.
25:24 Integrating Foundation Models: Trident incorporates 22 different deep learning models. The goal is to integrate these with Galaxy, allowing users to pull models directly from repositories like Hugging Face.
27:32 Use Case: Predicting Drug Response: The team is applying these models to TCGA breast cancer data to classify subtypes and aims to predict drug response in triple-negative breast cancer.
This presentation outlines analysis strategies for 10x Genomics Visium HD, a spatial transcriptomics technology with subcellular (2-micron) resolution.
32:53 Multiple Analysis Scales: Analysis can be performed at different resolutions: 8-micron bins, 2-micron bins, or at the individual cell level.
35:56 High-Resolution Clustering with fixture: The fixture tool enables high-resolution clustering on the full 2-micron data.
36:47 Cell-Level Analysis: The BintoCell tool segments cells from the H&E image and aggregates expression data to a cell-level matrix. Bangsue is then used for multi-sample clustering and batch correction.
41:45 Structure-Based Analysis of Spatial Omics Data
This talk introduces an analysis approach focusing on multicellular anatomical structures.
42:42 Example 1: Pancreatic Islets: Using imaging mass cytometry (IMC) data, the team reconstructs individual pancreatic islets to quantify their morphology.
44:42 Statistical Considerations: The speaker emphasizes using linear mixed-effect models to account for data correlation (e.g., multiple structures per slide) and avoid inflated false discovery rates.
49:12 The Spatula Package: The team developed the Spatula Bioconductor package to facilitate these types of structure-based analyses.
This work aims to speed up the manual curation bottleneck in the Vertebrate Genomes Project (VGP).
54:16 Curation with Hi-C Maps: Manual curation relies on Hi-C contact maps to identify misassemblies and organize scaffolds into chromosomes.
56:00 Galaxy Workflow for Curation: A Galaxy workflow produces a Hi-C map in Pretext format, overlaying it with evidence tracks like coverage, gaps, and gene annotations to provide context for curators.
56:44 Use Cases: The evidence helps identify breakpoints, distinguish gene-rich microchromosomes from assembly errors, and identify sex chromosomes.
59:53 Availability: The workflow is available in the IWC and is regularly updated for any diploid genome.